Anomalous Object Tracking in Distributed Camera Network

Abstract

The paper introduces a novel framework for real-time tracking of an object with higher precision in a Pan-Tilt-Zoom (PTZ) camera network. In the above-mentioned framework, the object which behaves anomalously is picked for tracking. An SVM classifier has been used to pick anomaly behaving object among all the objects present in a frame. It is a self-initializing algorithm as it does not require user intervention for object detection. Target detection is followed by its autonomous tracking in the distributed camera network. Hence it does not make the system bulky and takes less time to execute. The paper implements multiple object tracking using Discriminative Correlation Filtering with Channel and Spatial Reliability(CSRDCF) tracking algorithm. Existing works are based on Particle filtering or Kalman filtering algorithms which are computationally complex and not self-initializing.

Keywords

• object tracking
• self-initialization
• distributed camera network
• anomaly behaviour detection
• histogram comparison.